Inferring latent attributes of Twitter users with label regularization

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 185–195, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Inferring latent attributes of Twitter users with label regularization Ehsan Mohammady Ardehaly and Aron Culotta Department of Computer Science Illinois Institute of Technology Chicago, IL 60616 emohamm1@hawk.iit.edu, aculotta@iit.edu Abstract Inferring latent attributes of online users has many applications in public health, politics, and marketing. Most existing approaches rely on supervised learning algorithms, which re- quire manual data annotation and therefore are costly to develop and adapt over time. In this paper, we propose a lightly supervised approach based on label regularization to in- fer the age, ethnicity, and political orientation of Twitter users. Our approach learns from a heterogeneous collection of soft constraints derived from Census demographics, trends in baby names, and Twitter accounts that are em- blematic of class labels. To counteract the im- precision of such constraints, we compare sev- eral constraint selection algorithms that opti- mize classification accuracy on a tuning set. We find that using no user-annotated data, our approach is within 2% of a fully supervised baseline for three of four tasks. Using a small set of labeled data for tuning further improves accuracy on all tasks. 1 Introduction Data annotation is a key bottleneck in applying supervised machine learning to language process- ing problems. This is especially problematic in streaming settings such as social media, where mod- els quickly become dated as new linguistic pat- terns emerge. An attractive alternative is lightly supervised learning (Schapire et al., 2002; Jin and Liu, 2005; Chang et al., 2007; Grac¸a et al., 2007; Quadrianto et al., 2009; Mann and McCallum, 2010; Ganchev et al., 2010). In this approach, classifiers are trained from a set of domain-specific soft con- straints, rather than individually labeled instances. For example, label regularization (Mann and Mc- Callum, 2007; Grac¸a et al., 2007) uses prior knowl- edge of the expected label distribution to fit a model from large pools of unlabeled instances. Similarly, annotating features with their expected class fre- quency has proven to be an efficient way of boot- strapping from domain knowledge (Druck et al., 2009; Melville et al., 2009; Settles, 2011). In this paper we use lightly supervised learning to infer the age, ethnicity, and political orientation of Twitter users. Lightly supervised learning provides a natural method for incorporating the rich, declar- ative constraints available in social media. Our ap- proach pairs unlabeled Twitter data with constraints from county demographics, trends in first names, and exemplar Twitter accounts strongly associated with a class label. Prior applications of label regularization use a small number of highly-accurate constraints; for ex- ample, Mann and McCallum (2007) use a single constraint that is the true label proportions of an un- labeled dataset, and Ganchev and Das (2013) use cross-lingual constraints from aligned text. In con- trast, we use hundreds of constraints that are het- erogeneous, overlapping, and noisy. For example, we constrain the predicted …

Content truncated to 500 words